Topics:
In 2017 we will be encouraging submissions on the
Special Topic of Monitoring and Measurements in Experiments:
- Use of system monitoring data to extract meaningful information from experiments such as those that seek to assess resource contention, determine root causes of performance variations, or understand system failures and fault propagation
- Proper design of experiments, including measurement procedures and evaluation
- Experimental rigour and reproducibility.
- Inclusion of measurements of relevant ecosystem information, such as encompassing system variables, concurrent job mix, and system environmental data (e.g., facilities)
- Publication of relevant datasets in support of validation and additional analysis
In addition to general topic areas, including, but not limited to:
Data collection, transport, and storage
- Monitoring methodologies and results for all HPC system components and support infrastructure (e.g., compute, network, storage, power, facilities)
- Design of systems and frameworks for HPC monitoring which address HPC requirements such as:
- Extreme scalability
- Run time data collection and transport
- Analysis on actionable timescales
- Feedback on actionable timescales
- Minimal application impact
- Extraction and evaluation of resource utilization and state information from current and next generation components
Analysis of large-scale data and system information
- Extraction of meaningful information from raw data, such as system and resource health, contention, or bottlenecks
- Methodologies and applications of analysis algorithms on large scale HPC system data
- Visualization techniques for large scale data (addressing size, timescales, presentation within a meaningful context)
- Evaluation of correlative relationships between system state and application performance via use of monitored system data
Response to and utilization of processed data and system information
- Mechanisms for feedback and response to applications and system software (e.g., informing schedulers, down-clocking CPUs)
- HPC application design and implementation that take advantage of monitored system data (e.g., dynamic task placement or rank-to-core mapping)
- System-level and Job-level feedback and responses to monitored system data
- Job scheduling and allocation based on monitored system information (e.g. contention for storage or network resources)
- Integration of system and facilities data for system and site operational decisions
- Use of monitored system data for evaluation of future systems specifications and requirements
- Use of monitored system data for validation of systems simulations
Experience reports and System Operations
- Design and implementation of monitoring tools as part of HPC operations
- Experiences with monitoring and analysis methodologies and tools in HPC applications
- Note this is not meant to include application performance analysis tools such as open|speedshop or craypat
- Experiences with monitoring and analysis tools for HPC systems specification/selection
- Sub-optimal approaches taken because there currently isn’t another way (include associated gap analysis)
- How not to do it, with explanations, benchmarks, or analysis of code to save the rest of us from trying it again